Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1998-06-26
2001-04-17
Hudspeth, David (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C379S088180
Reexamination Certificate
active
06219643
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to the field of natural language recognition systems. More particularly, this invention relates to the field of analyzing dialog and dialog state data in a natural language recognition system.
BACKGROUND OF THE INVENTION
Interactive telephone systems are used to provide users access to a variety of services without the necessity of requiring a live operator to interface with the user. Initially such systems were of the type that played an audio message and then reacted upon sensing the user's response via entering one or more key strokes on a touch tone telephone keypad to generate well known dtmf tones. It is well recognized that such systems are not preferred for a variety of reasons including that they are cumbersome to use and further because this type of response is not natural for most users. Particularly, it is undesirable for a user to listen to a series of response choices and determine which of the ‘programmed responses’ is most appropriate while concurrently remembering which of the several keystrokes must be used to elicit that response.
With the advent of natural language recognition systems, users could respond to interactive telephone systems using more natural spoken responses. Such systems are used for a variety of applications. One known example is for providing information and services regarding flight availability, flight times, invoicing, payments and flight reservations and the like for a predetermined airline. Another well known use for such systems includes gaining information regarding stocks, bonds and other securities, purchasing and selling such securities, and gaining information regarding a user's stock account. Also, systems exist for controlling transactions in accounts at a bank. Other applications are also used.
FIG. 1
shows a finite state diagram of an exemplary prior art natural language interactive telephone system for an airline application such as that set forth above. In state
100
the application receives an incoming telephone call. This is being welcome in which the system plays a greeting to the caller during which the system identifies itself to the user. The state
102
represents the main menu of the airline interactive telephone system. During this state the system asks the user which of its several modes the user wishes to invoke. For example, the user could choose to obtain flight scheduling information in a first mode
104
. Alternatively, the user could choose to learn about availability of seats for given flights in a second mode
106
. Also, the user could choose to learn about actual arrival times for in-the-air flights in a third mode
108
. The user can also use the system to purchase ticket in state
110
. Additionally, the user could obtain information regarding the cost of flying between two cities in the state
112
. Such a system could be configured to have many other modes for achieving other desired functions.
Any typical system each of the mode states
104
through
112
will cause the system to traverse through a predetermined series of states. To avoid obscuring the invention in extraneous details of which are unrelated to the principal elements of this invention, only one of the possible series of states are shown herein. In this example, a sample series of states are shown for purchasing a ticket. In the state
114
the system queries the user regarding the city in which the flight is to begin. In the state
116
the system queries of the user regarding the city in which the flight is to terminate. The system then queries the user to learn when the user wants to travel in the state
118
. The state
118
will be discussed in more detail below. Upon determining these facts, the system can then access the database of flight information in the state
120
present the list of relevant flights to the user. In the state
122
the user selects a flight from among the several choices. Thereafter, the user can exit the system or return to a previously described portion of the state diagram to perform another operation.
Depending upon the complexity of the system design, the user may be required to provide each piece of information sequentially, as described in the example above, or may be allowed to provide all the pieces of information concurrently in a single dialog state transaction. Thus the finite state diagram of
FIG. 1
could be shown with more or fewer number of finite states for achieving the same function. This relative system complexity is known in the art and is ancillary to the present invention.
Conventionally, a single information transaction comprising a single utterance by the system and then a single response utterance by the user is termed a ‘dialog state.’ Several transfers of information related to a particular topic and carried on between the system and the user via the telephone interface port is termed a ‘dialog’ in the state-of-the-art. Generally a dialog includes several dialog states. A telephone call includes all the dialogs and dialog states which occur between a system and a user from the time the equipment goes off-hook until it returns to an on-hook state.
According to usual practice, upon receiving a voice communication from user within a dialog state the system undertakes two principal operations in response thereto. First, the system performs a natural language speech recognition operation to determine the words that were spoken by the user. To aid in the recognition operation, the system is programmed with a predetermined set of anticipated responses. A nonsequitor response will generally be unrecognizable as “out of grammar.” For example, if the system queried the user about the destination city for flight and the user responded “7 p.m. on Thursday”, the system will likely not be programmed to recognize those words in the context of this dialog state. Second, the system must determine whether it ‘understands’ the words to be recognized in the context of the anticipated dialog state.
It will be understood by persons of ordinary skill in the art that establishing proper dialog state interactions is a demanding problem for natural language speech recognition systems. Consider for example of the main menu state
102
. The system utterance could be “what do you want to do?” In response, the user could say “I need to visit my Aunt Gertrude who's sick in the hospital as soon as possible.” The result of such a dialog state would not likely provide any useful information.
In the alternative, the system utterance could be “do you want to obtain flight information, flight availability, arrival times, purchase ticket, obtain cost information, or . . . ?” This system utterance is far more likely to receive a user utterance reply which it can recognize and understand. Thus, if the user replies “purchase ticket” the system will understand what the user wants to do. On the other hand, if the user replies “book a flight”, “reserve a ticket”, “make a reservation” or other similar utterances the system may or may not understand the user's intent. The main menu dialog can be designed to allow for recognizing and understanding a variety of user utterance replies to a specific system utterance.
Due to idiomatic nature of natural language it will be readily understood that some users will not understand a particular system utterance and similarly, some systems will not understand a particular user utterance. Generally speaking when a user fails to understand a system utterance, the user will either hang up or respond with a user utterance that the system views as nonsequitor. In either case the subject dialog state failed to produce the intended result. Consider the state
118
of FIG.
1
. That example indicates that the system queries the user for when travel is desired. It may be that the system wants to know time of day, day of the week or calendar date or some combination or all of these elements. A user could easily answer with the wrong, or unexpected information. This would be viewed as a failed dialog state.
The res
Cohen Michael H.
Jackson Eric G
Kurada Vijaya L
Haverstock & Owens LLP
Hudspeth David
Lerner Martin
Nuance Communications Inc.
LandOfFree
Method of analyzing dialogs in a natural language speech... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of analyzing dialogs in a natural language speech..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of analyzing dialogs in a natural language speech... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2546778